Representation Learning on Graphs with Jumping Knowledge Networks, Keyulu Xu, Chengtao Li, Yonglong Tian, Tomohiro Sonobe, Ken-ichi Kawarabayashi, Stefanie Jegelka, 2018Proceedings of the 35th International Conference on Machine Learning (ICML), Vol. 80 (PMLR) - Introduces Jumping Knowledge Networks, an architecture that combines information from different layer depths to prevent oversmoothing and improve representation quality.
DeepGCNs: Can GCNs Go as Deep as CNNs?, Guohao Li, Chenxin Xu, Weigang Zhang, Zhiwu Lu, 2019Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) (IEEE)DOI: 10.1109/ICCV.2019.00194 - Proposes a framework for building deep GCNs using techniques like residual connections, achieving state-of-the-art results and demonstrating the feasibility of deeper GNNs.
PairNorm: Tackling Oversmoothing in GNNs, Lingxiao Zhao, Leman Akoglu, 2020Proceedings of the 29th International Joint Conference on Artificial Intelligence (IJCAI)DOI: 10.24963/ijcai.2020/220 - Introduces PairNorm, a graph-specific normalization technique designed to explicitly address the oversmoothing problem by controlling the variance of node embeddings.